Kenji KITA
A Study on Language Modeling
for Speech Recognition
Abstract:This report is concerned mainly with language modeling for automatic speech recognition.
Speech recognition based on acoustic information alone does not permit good performance for
large-vocabulary tasks. Linguistic information can assist in solving speech recognition problems.
The report proposes efficient speech recognition algorithms that take full advantage of the
grammatical constraints supplied by syntactic language models, and introduces some stochastic
language models to improve speech recognition performance.